SinoLC-1:中国首幅1米分辨率土地覆盖图数据集

SinoLC-1：中国1米分辨率土地覆盖图为首个具有中国国家尺度覆盖，空间分辨率1米的土地覆盖专题图。针对大范围高分辨率土地覆盖制图中地物复杂多样、高精度训练样本缺乏、制图方法区域迁移性要求高等关键难题，中国地质大学（武汉）张洪艳教授团队于2022年10月提出了一种基于弱监督深度学习的跨分辨率大规模土地覆盖制图网络（L2HNet），利用低分辨率土地覆盖图作为训练集，实现了高分辨率高精度土地覆盖制图。该方法摆脱了高分辨率土地覆盖制图任务对大规模人工手动标注训练集的依赖，大幅减少了大范围高分辨率土地覆盖制图的人力与时间成本，作为阶段性研究进展发表在遥感科学领域权威期刊《ISPRS Journal of Photogrammetry & Remote Sensing》上（链接：https://doi.org/10.1016/j.isprsjprs.2022.08.008）。在此基础上，团队进一步构建了中国全域土地覆盖制图框架，利用现有三种10米土地覆盖图（FROM_GLC10、ESA_WorldCover和ESRI_GLC）、开源道路图（Open Street Map，OSM）数据作为训练标签，基于1米分辨率Google开源遥感数据开展全国高分辨率土地覆盖制图，并采用10万目视解译参考点和第三次全国土地调查数据进行精度验证。最终，用时10个月处理约73TB高分辨率多源数据，成功研制了我国第一幅1米分辨率土地覆盖一张图（SinoLC-1）。目前，该专题图描述论文已在地球科学数据权威期刊《Earth System Science Data》上发表（链接：https://essd.copernicus.org/preprints/essd-2023-87/）。

SinoLC-1: The 1-meter resolution national land cover map is the first nationwide land cover thematic dataset with a spatial resolution of 1 meter. To address the core challenges in large-scale high-resolution land cover mapping, including diverse and complex ground features, scarcity of high-quality training samples, and high demand for regional transferability of mapping methodologies, the team led by Prof. Zhang Hongyan from China University of Geosciences (Wuhan) proposed a cross-resolution large-scale land cover mapping network (L2HNet) based on weakly supervised deep learning in October 2022. This method uses low-resolution land cover maps as training datasets to enable high-resolution and high-precision land cover mapping, eliminating the dependence of high-resolution land cover mapping tasks on large-scale manually annotated training datasets and significantly reducing the labor and time costs associated with large-scale high-resolution land cover mapping. This staged research outcome was published in the authoritative remote sensing science journal *ISPRS Journal of Photogrammetry & Remote Sensing* (DOI: https://doi.org/10.1016/j.isprsjprs.2022.08.008). Building on this work, the team further constructed a nationwide land cover mapping framework. They utilized three existing 10-meter resolution land cover products (FROM_GLC10, ESA_WorldCover, and ESRI_GLC) and open-source road map data (Open Street Map, OSM) as training labels, and conducted nationwide high-resolution land cover mapping using 1-meter resolution Google open-source remote sensing data. The mapping accuracy was validated using 100,000 visually interpreted reference points and data from the Third National Land Survey. Ultimately, after processing approximately 73 terabytes of high-resolution multi-source remote sensing data over 10 months, the team successfully developed the first nationwide 1-meter resolution land cover map (SinoLC-1) for China. Currently, the paper describing this thematic map has been published in the authoritative Earth science data journal *Earth System Science Data* (preprint link: https://essd.copernicus.org/preprints/essd-2023-87/).

搜集汇总

背景与挑战

背景概述

SinoLC-1是中国首个1米分辨率全国尺度土地覆盖图数据集，覆盖2002至2021年，基于深度学习框架L2HNet和开放数据生成，整体精度达73.61%。该数据集通过弱监督学习方法，显著降低了大规模高分辨率土地覆盖制图的人工和时间成本，适用于遥感与测绘领域的研究和应用。